AI News · August 15, 2026

llama.cpp adds MiniMax as GLM 5.2 price jumps 205%

238,868 stars today64 new projects+6,339 speech-to-speech

Z.AI raised the input price for GLM 5.2 to $1.19 per 1M tokens, a 205 percent increase. The same vendor raised GLM 5.1 by 47 percent and GLM 4.6 by 10 percent, while cutting GLM 5 by 37 percent. Moonshot AI raised Kimi K2.7 Code input price by 6 percent.

Alibaba cut input prices across three Qwen models, with the largest cut at 22 percent for Qwen3.5 397B A17B. Qwen3.5-35B-A3B fell 12 percent and Qwen3 VL 30B A3B Instruct fell 13 percent. The moves leave Z.AI and Moonshot on the high end and Alibaba on the low end for input pricing.

On the release side, llama.cpp b10437 added support for MiniMax-Text-01 and MiniMax-M1. Pydantic AI v2.31.0, XGBoost v3.4.1, AG2 v1.0.2, Open Interpreter rust-v0.0.38 and LoopX v0.4.7 also shipped. LoopX v0.4.7 focuses on longer goal loops and managed turn connectors for DeepSeek Harness.

The ecosystem gained 238,868 GitHub stars across tracked projects today, with speech-to-speech leading at 6,339 stars. Sixty-four new projects entered the index. This week, 1,962 new projects and 100 Hugging Face spaces were added.

Releases of the day

llama.cpp b10437 added support for MiniMax-Text-01 and MiniMax-M1. Pydantic AI v2.31.0 lets developers build a UIEventStream without a run_input and gives AGUIEventStream its own thread_id and run_id. XGBoost v3.4.1 fixed batch prediction for SparseVector features and model slicing with category containers. AG2 v1.0.2 shipped a robots.txt fix to stop advertising the /latest archive and added documentation for the AG2 Code Assistant. Open Interpreter rust-v0.0.38 is a minor release. LoopX v0.4.7 keeps long-range goal loops continuous, connects DeepSeek Harness through a managed Turn connector, and adds per-goal handoff with claim and lease permissions. For users, these changes mostly improve stability and agent loop control.

Pricing moves

Input prices moved in two directions today. Z.AI raised GLM 5.2 by 205 percent to $1.19, GLM 5.1 by 47 percent to $1.40, and GLM 4.6 by 10 percent to $0.55 per 1M tokens. Z.AI also cut GLM 5 by 37 percent to $0.60. Moonshot AI raised Kimi K2.7 Code by 6 percent to $0.71. Alibaba cut Qwen3.5-35B-A3B by 12 percent to $0.22, Qwen3.5 397B A17B by 22 percent to $0.39, and Qwen3 VL 30B A3B Instruct by 13 percent to $0.13. The largest single move was the GLM 5.2 increase.

▲ 205%GLM 5.2Z.AI · input $0.39 → $1.19 per 1M tokens
▲ 47%GLM 5.1Z.AI · input $0.95 → $1.4 per 1M tokens
▲ 6%Kimi K2.7 CodeMoonshot AI · input $0.67 → $0.71 per 1M tokens
▼ 37%GLM 5Z.AI · input $0.95 → $0.6 per 1M tokens
▼ 12%Qwen3.5-35B-A3BAlibaba · input $0.25 → $0.22 per 1M tokens
▼ 22%Qwen3.5 397B A17BAlibaba · input $0.5 → $0.39 per 1M tokens
▲ 10%GLM 4.6Z.AI · input $0.5 → $0.55 per 1M tokens
▼ 13%Qwen3 VL 30B A3B InstructAlibaba · input $0.15 → $0.13 per 1M tokens

Emerging projects

Eight new projects entered the index. KiroCrew is a persistent workspace for development work that self-improves beyond one session. doc7 turns documents into AI-ready Markdown with visual understanding

Research pick

Source: olud.ai tracking of 10,000+ open-source AI projects, 300+ models and live provider pricing. Figures are measured, not estimated. All releases · Live pricing · Latest in AI

← All AI news